@Szabolcs Horvát @Andrew Raffensperger you may find this interesting, I've just submitted a function repository function to correctly register a format's extension and mime type. It can be found here.

Thank you, this is very valuable!

Hey Sean, thanks for taking the time to respond.

Here is working code that can be called after Quit[] for:

1. 11.3.0 for Microsoft Windows (64-bit) (March 7, 2018)
2. 10.1.0 for Microsoft Windows (64-bit) (March 24, 2015)
3. 11.3.0 for macOS (I'm running 12.0.0 ATM so I can't get the old version string.)

Module[{name="ABC",exts={"*.abc"}},Quiet@Import["__FAKE FILE__",""];
ImportExport`RegisterExport[name,Function[Print[{"export args:",##}];Export[#1,#2,"Byte"]],"Extensions"->exts];
ImportExport`RegisterImport[name,Function[Print[{"import args:",##}];Import[#1,"Byte"]],"Extensions"->exts];
If[!MemberQ[FileFormatDump`$FILEFORMATS,name],AppendTo[FileFormatDump`$FILEFORMATS,name]];
FileFormatDump`$FILEFORMATMATRIX[name]={name,True,False,False,False,False,exts,{},None,{}};
file=CreateFile[]<>".abc";
Print[file];
out={1,2,3};
Export[file,out];
Import[file]===out]

This code errors in 12.0.0.

I understand this is undocumented code so I kept my post kinda vague. I was more concerned that code executed in the front end depends on the liveliness of the associated kernel. This was just the first example I found where I got bit by this behavior.

Thanks for the response.

Maybe I wasn't clear enough with my example. My code is NOT running as part of init.m or any related initialization. I am simply talking about the first cell in a notebook.

And the bug is: depending on whether the kernel has been started or not, evaluating code in the first cell produces different results. This is insane!

Why do I have to type something into a blank cell, evaluate it, wait for the kernel to return, and then evaluate the rest of my notebook?